An Evaluation of Duplicate Adverse Event Reports Characteristics in the Food and Drug Administration Adverse Event Reporting System.

Abstract

Introduction: The Food and Drug Administration (FDA) Adverse Event Reporting System (FAERS) receives duplicate reports of adverse events associated with drug and therapeutic biological products. Duplicate reports, defined as multiple reports of the same adverse event(s) related to the administration of the same marketed product(s) to the same individual patient at a particular point in time, may be received in FAERS for many reasons. The presence of duplicate reports can negatively impact public health surveillance efforts by impeding both safety signal identification and signal evaluation.

Objectives: To characterize the features and contributing factors associated with duplicate reports in FAERS.

Methods: We manually assessed a convenience sample of individual case safety reports (ICSRs) for duplication, resulting in two data sets: one consisting of non-duplicate reports and one with duplicate reports. We then compared key features of these two datasets, including both structured and unstructured data fields. Key comparison features included: report and reporter type, country of report origin, data source for report, and outcome. In addition, we evaluated information similarity of reports for seven data elements (e.g., age, sex, suspect products) within sets of duplicates using both structured and unstructured fields. We used pairwise sentence bidirectional encoder representations from transformers (SBERT) cosine similarity scores to examine free-text narrative similarity.

Results: Among the 2297 reports in the sample, 272 (12%) were classified as duplicates, consisting of 85 unique duplicate sets. Compared with non-duplicate reports, duplicates were more likely to be foreign reports (85% versus 51%), reported by healthcare professionals (89% versus 74%), mention other regulatory authority databases (44% versus 20%), or describe published case reports (28% versus 19%) (p = 0.01). Within sets of duplicates (n = 85), coded information was frequently different, with only 16% (n = 14) having concordance of all 7 data elements. The narrative was highly similar among most sets of duplicates; we found that the median similarity score for the duplicate pairs was 0.74 compared with 0.38 for non-duplicate pairs.

Conclusions: We observed differences in the attributes of and potential contributors to duplicate reports in FAERS that may inform duplicate prevention, detection, and management strategies. However, further studies are needed to better understand the implications of these findings and how potential regulatory changes and technological advances can be leveraged to further address duplicate reporting in adverse event reporting systems.